[2021-01-08] Prof. Hung-Wei Tseng, University of California, Riverside, "Efficient Use of AI Accelerators"

Poster:Post date:2020-12-30
Title: Efficient Use of AI Accelerators
Date: 2021-01-08 3:40pm-5:00pm
Location: R102, CSIE
Speaker: Prof. Hung-Wei Tseng, University of California, Riverside
Hosted by: Prof. Chia-Lin Yang


The significance of artificial intelligence (AI) and machine learning (ML) applications has changed the landscape of computer systems: AI accelerators start to emerge in a wide range of devices, from mobile phones to data center servers. With specialized matrix units (MXUs), these accelerators offer orders of magnitude performance gain or energy savings for the core computation in AI/ML workloads. However, the benefits do not come without overhead, as we need to additionally reshape, quantize, de-quantize, move in/out data for these MXUs.
To identify the overhead of “non-computational” part of training in cloud infrastructures, we developed a profiling tool for Google’s TPU-based cloud servers. We have identified that cloud TPUs are under-utilized because of the aforementioned overhead. We developed VarifocalStorage to demonstrate the potential of using intelligent storage in preparing datasets for these AI/ML accelerators. This talk will also demonstrate the potential of widening the applications of AI/ML applications and discuss the future of these accelerator-based architectures.
Hung-Wei is currently an assistant professor in the Department of Electrical and Computer Engineering at the University of California, Riverside. He is now leading the Extreme Storage & Computer Architecture Laboratory and focusing on tackling the performance issues in modern heterogeneous computer systems through intelligent data storage. He is recognized by facebook faculty research award and IEEE Micro "Top Picks from Computer Architecture" in 2020 for his research in accelerating data-intensive applications through revisiting the storage system design. He got his PhD from the Department of Computer Science and Engineering at the University of California, San Diego. 
Last modification time:2020-12-30 AM 9:05

cron web_use_log